A regression approach to the analysis of data arising from cluster randomization.

A generalized least squares regression approach is proposed for the analysis of data arising from experimental studies involving cluster randomization and non-experimental studies in which the major treatment factor corresponds to a characteristic which applies at the cluster level. This approach is more flexible than that provided by the analysis of variance, and unlike ordinary least squares, provides significance levels which are adjusted for the correlation among elements within the same cluster. Two examples are presented.